Probabilistic analysis of crack growth in railway axles using a Gaussian process

Abstract

To reduce maintenance costs, it is important to carry out probabilistic analyses on railway vehicle components. In this work, a data-driven approach based on a Gaussian process for regression is developed to determine the probability of axle failure caused by crack growth in railway axles. For complicated failure modes, it is difficult or even impossible to build a reliable analytical or simulation model before using an analytical approach. The main purpose of this work is to develop an algorithm to infer the distribution of crack growth from limited measured data without having to build an underlying model. The results of the case study show that the determined timing for the first inspection and the probability of failure coincide with the known results derived by analytically based approaches. The problems associated with modelling and calibration can be overcome by a data-driven approach. The developed Gaussian process model can serve as a complementary instrument to validate other analytically based approaches or numerical analyses. The model can also be applied to the probabilistic analyses of other railway components.